문제
The softmax function turns a list of scores (logits) into probabilities. Language models use it to choose the next word:
softmax(x)_i = e^(x_i) / (e^(x_1) + e^(x_2) + … + e^(x_n))
Large scores make e^x overflow, so first subtract the largest score from every score. The result is mathematically the same.
Implement softmax(logits).
Input
logits: A list of scores.
Output
Return the list of probabilities, in the same order. They are all between 0 and 1 and add up to 1.
제약 조건
1 ≤ length ≤ 1,000
-1,000 ≤ each score ≤ 1,000
예시
입력:
logits = [1.0, 2.0, 3.0]
예상 출력:
expected = [0.090031, 0.244728, 0.665241]
입력:
std::vector<double> logits = {1.0, 2.0, 3.0};
예상 출력:
std::vector<double> expected = {0.090031, 0.244728, 0.665241};
입력:
double[] logits = new double[] {1.0, 2.0, 3.0};
예상 출력:
double[] expected = new double[] {0.090031, 0.244728, 0.665241};
설명
e¹ ≈ 2.718, e² ≈ 7.389 and e³ ≈ 20.086 add up to 30.193, so the probabilities are about [0.090031, 0.244728, 0.665241].
입력:
logits = [1000.0, 1000.0]
예상 출력:
expected = [0.5, 0.5]
입력:
std::vector<double> logits = {1000.0, 1000.0};
예상 출력:
std::vector<double> expected = {0.5, 0.5};
입력:
double[] logits = new double[] {1000.0, 1000.0};
예상 출력:
double[] expected = new double[] {0.5, 0.5};
설명
Computing e^1000 directly overflows. Subtracting the maximum gives [0, 0], and e⁰ / (e⁰ + e⁰) = 0.5 for both.
접근 방법
먼저 스스로 풀어 본 뒤 힌트를 하나씩 열어 보세요.
힌트 1
Subtract the largest logit before exponentiating so the largest transformed score becomes zero and overflow is avoided.
힌트 2
Compute each exponent, sum them once, then divide every exponent by that total to get probabilities.
힌트 3
Make sure the returned values stay in the same order as the input and still sum to one after normalization.
예상 복잡도
O(n) time, O(n) extra space
참고 자료
이 문제는 CodeUtility에 맞게 수정되었습니다. 위의 설명과 제한 조건을 따르세요.